Machine learning for biological inference across spatial, multimodal, and clinical dimensions in precision medicine

This dissertation develops machine learning methods to address key barriers in precision oncology, focusing on spatial biology, multimodal data integration, and clinical outcome prediction. It introduces a spatially informed imputation strategy for multiplex tissue imaging, a low-complexity embedding aggregation framework for integrating diverse biomedical data, and interpretable models to identify predictors of immunotherapy response in melanoma. Together, these approaches improve data quality, enable interpretable integration of high-dimensional data, and support biologically and clinically meaningful inference in cancer research.